CSTR comes of age

The Centre for Speech Technology Research (CSTR) is twenty-one this year, and is holding a two-day celebration on Thursday and Friday, September 1st and 2nd.
Founded in 1984, CSTR is concerned with research in all areas of speech technology including speech recognition, speech synthesis, speech signal processing, information access, multimodal interfaces and dialogue systems. We have many collaborations with the wider community of researchers in language, cognition and machine learning for which Edinburgh is renowned.

Simon Stringer Abstract

A dynamical systems approach to modelling brain function:
Embedding biophysically accurate neural network models in 3D virtual reality environments.

The power of modern computers now makes it possible to investigate the dynamics of neural processing in the brain through computer simulations of biophysically accurate neural network models embedded and self-organising within realistic 3D virtual environments. Two areas of brain function which are immediate candidates for this approach are object recognition in the ventral visual processing stream, and the representation of space in the hippocampus and related brain regions.

In self-organising neural network models of object recognition in the ventral visual stream, the spatiotemporal statistics of how objects move in the world are critical to how the network self-organises during learning, and whether transform invariant representations develop. Because of this, we have begun to use visual input from 3D virtual reality environments, in which the time evolution of the positions and velocities of visual stimuli can be carefully controlled, as well as the point of view of the network or agent within the environment.

Further work needs to be done to improve the biophysical accuracy of existing vision models, which typically rely on rate-coded Hebbian learning rules, and which form associations between all neurons which are co-active. Because of this limitation, current models are usually trained with only a single stimulus at a time. The incorporation of more accurate biophysical learning mechanisms, such as spike time dependent learning, may play an important role in enabling neural network models to process visual input from complex scenes with multiple features and objects.

Investigations of how the brain learns to represent 3D space and perform spatial navigation are also set to benefit from the use of 3D virtual reality simulations. For example, 3D virtual reality tools should have an important role to play in simulating the development of "place cells" found in the rat hippocampus, which appear to underpin the animal's ability to navigate. A key question is how do place cells develop their firing properties using low-level sensory inputs such as vision and idiothetic (self-motion) signals to guide the self-organisation of the synaptic connections. As above, further improvements in biophysical accuracy may be required in order to allow these models to process realistic visual input from complex environments.

Rasmus Petersen Abstract

Neuronal Computation in Somatosensory Cortex

What are the principles that underlie sensory information processing in the mammalian brain? As a model preparation for approaching this issue, I have studied how rats use their vibrissae to sense their environment. The vibrissae circuits of the rodent brain have a relatively simple, modular structure that make it a very useful system for studying neural computing.

Although the way that neurons respond to simple vibrissa stimuli (deflections of individual vibrissae repeated ca. once per second) is well understood, these stimuli are artificial. In natural rat behaviour, vibrissae vibrate in a much more complex way. An important open question is therefore how complex, naturalistic stimuli are encoded, and what computations neurons perform on them. To address this, I used a systems identification approach. I used a computer-controlled piezoelectric actuator to move the vibrissae of an anaesthetised rat in the vertical plane. I recorded the spikes that individual neurons from primary somatosensory cortex (S1) fired in response with a microelectrode. The standard technique (spike-triggered averaging, STA) used to identify a neuron’s stimulus-response relationship is to apply a white noise stimulus (here, vibrissa position, band-limited 0-150Hz), record the consequent spike times and cross-correlate the spike train with the stimulus. Provided that the neuronal firing rate can be modelled as convolution of the stimulus with a given filter, the cross-correlation function estimates that filter.

However, when I applied this approach to neurons in S1, STA did not yield a good model of the neuronal response. I therefore applied a new, mathematically more powerful technique (spike-triggered covariance, STC). Unlike STA, STC does not assume a linear stimulus-response relationship. In STC, neuronal firing rate is modelled as convolution of the stimulus with a set of orthogonal filters, the results of which are combined via a non-linear function. Both filters and non-linearity can be estimated from white noise data. I found that S1 neurons could be well-described by multiple filters of two types. (1) Neurons were excited by vibrissa velocity/acceleration on a time-scale of ~30 ms. (2) An unexpected result was that neurons were also inhibited by vibrissa dynamics on a longer (~150 ms) time-scale. Spike production thus reflects an interplay between short-time scale stimulus dynamics and longer time-scale “context”. The significance of this study is that it quantifies, for the first time, the computation that S1 neurons perform on a complex, naturalistic tactile stimulus. An aim for future research is to use modelling to probe how the computation is implemented by neural circuits.

Jan Karbowski Abstract

Computational Systems biology and locomotion of C. elegans worms

In the beginning of the talk I will present briefly my current research interests in computational biology. For the rest of the talk, I will focus on one particular topic: locomotion of C. elegans worms.

C. elegans are standard biological organisms for molecular and genetic studies. Their behavioural repertoire is quite limited and sinusoidal locomotion plays a major part in it. Despite the identification of hundreds of genes involved in C. elegans locomotion, we do not yet have an understanding of its control. In particular, molecular, neural, and network level mechanisms are largely unknown. I will present a computational, system-level, model of C. elegans sinusoidal (undulatory) locomotion and provide its experimental support. The model is composed of two parts: microscopic (neuromuscular-sensory circuit) and macroscopic (biomechanics). The microscopic part reveals how oscillatory activity, used in locomotion, is generated. The macroscopic part explains the emergence of conserved relationships between mechanical parameters observed experimentally.

Irina Erchova Abstract

Reverse engineering neuronal signal transmissions

In the nervous system signals from the external world are converted into a spatially distributed series of electrical pulses emitted by individual neurons. Understanding the mechanisms of signal transmission, filtering and targeting are the keys to develop effective tools for diagnosing neuronal pathologies and enabling the development of brain – computer interfaces (BCIs).

My research focuses on signal coding and the mechanisms implicated in the propagation and direction of information flow in complex networks. The term ‘reverse engineering’ implies a systematic methodology by which quantitative mathematical models of brain function can be developed, based on measured data and validation of models in prediction-driven experiments.

I am going to show how the intrinsic properties of cells and their local connectivities might predict optimal signal coding in the system. I will discuss the origin of the time scales for neural information representation and relate them to underlying biophysical parameters. In addition, I will show how response to a stimulus depends on ongoing network activity. This activity reflects, for example, previous experience, the current environment, or specific attention. Network modulation gates information flows and is thought to promote experience dependent changes in neuronal activity. I will illustrate a role of a local network modulation on plasticity in context-dependent stimuli integration in sensory cortex and discuss a possible role of global dynamic network modulation (brain rhythms) in relation to signal propagation and formation of memories.

Jeremy Caplan Abstract

Linking brain activity and behaviour:
Memory for associations and lists

Memory for structured information like pairs of items (e.g., SALT-PEPPER) or lists (e.g., the alphabet) are critical for human functioning. Understanding the behavioural and neural basis of such function requires multiple approaches.

I present one path through an integrative research programme. In particular, behavioural data and cognitive modelling suggested that memory for both associations and lists rely on the same cognitive processes. Electroencephalographic brain activity data recorded during this behaviour lends further support to this notion when viewed by conventional analysis methods. However, a multivariate analysis of the brain-behaviour correlation reveals study-related activity that appears specialized for list memory. Thus, cognitive neurophysiology findings challenge cognitive theory in ways that were not evident with a purely behavioural approach.

A separate study connects associative memory in rats to associative memory in humans. Human brain activity as measured with functional magnetic resonance imaging reveals two distinct functional networks that allow participants to learn conflicting information but each relying on a different neuromodulatory system. One functional network involves the basal forebrain, paralleling findings from rats.

These examples illustrate how a research programme linking brain activity and behavioural models and experiments can not only inform the component fields of study, but the process of connecting disparate fields itself can lead to insights into brain activity, behaviour and the human experience.

Neuroinformatics Talks

Thursday, 1 September 2005
Jim Howe Lecture Room, Forrest Hill

1.30 p.m. Dr. Jeremy Caplan, Toronto

Linking brain activity and behaviour:
Memory for associations and lists.

2.15 p.m. Dr. Irina Erchova, Edinburgh

Reverse engineering neuronal signal transmissions

3.00 p.m. – 3.15 p.m. COFFEE

3.15 p.m. Dr. Jan Karbowski, California Institute of Technology

Computational Systems biology and locomotion of C. elegans worms.

4.00 p.m. Dr. Rasmus Petersen, Manchester

Neuronal Computation in Somatosensory Cortex

4.45 p.m. Dr. Simon M. Stringer, Oxford

A dynamical systems approach to modelling brain function:
Embedding biophysically accurate neural network models in 3-D virtual reality environments.

5.30 p.m. Feedback session for AT staff

Informatics News

Congratulations to Jane Hillston, Taku Komura and Sethu Vijayakumar - see earlier entries for details.
In an effort to improve communication within the School, I've set up this experimental Informatics News blog. For the time being this is hosted at
http://homepages.inf.ed.ac.uk/mfourman/blogs/news/
with an atom feed at
http://homepages.inf.ed.ac.uk/mfourman/blogs/news/atom.xml
You should be able to use this using Mozilla Firefox Live Bookmarks
These URLs will almost certainly change in the near future - any changes will be advertised here.
Information on how to submit news items for the blog will be added soon.

Comments are welcome - remember this is a public forum.

Taku Komura appointed to a Lectureship

Taku Komura will join IPAB in February 2006

Taku received his received his PhD (2000), MSc (1997) and BSc (1995) in Information Science from the University of Tokyo. He worked as a postdoctoral researcher at RIKEN, Japan to simulate various human motion using the musculoskeletal model.
He is currently an Assistant Professor in the Department of Computer Engineering and Information Technology at City University of Hong Kong.
His work spans computer graphics, robotics and biomechanics, and his interests include human motion analysis and synthesis, physically-based animation and real-time computer graphics.

Jane Hillston Awarded EPSRC Advanced Fellowship

Process Algebra Approaches to Collective Dynamics

1 October 2005 - 30 September 2010

The aim of this scientific research programme is to help to understand how we can predict the dynamic behaviour of a population when we can model the behaviour of the individuals.
There are application areas in many scientific fields for a question such as this, but one of the most important is in biology. Biologists study cells, reagents, molecules, and molecular complexes in order to gain insights into growth and change. Here carefully defined models have increasing importance because more and more biological research is now being done using computer-based analysis. That form of research complements the biological research which is traditional laboratory science, based on controlled experiments, data gathering, and observation. Models allow the biologists to try to make sense of the data they have observed, forming hypotheses which may then be tested in further experiments. Their research findings inspire the design of new drugs and treatments which fight illness and improve human and animal health.
Scientists need languages in order to communicate effectively and unambiguously. The models developed by biologists are often written in the mathematical notation for calculus invented by Leibniz, as differential equations. Whilst this notation has many virtues it is not always the most intuitive to use, and can be far removed from the cellular processes being described. So biologists have been investigating the use of other system description languages, from other areas of science.
Computer science has more experience than the other sciences in formal language design. In computer science formal languages have been used for a variety of purposes, one of which is the design and validation of programs. One class of languages which have been used for this purpose are process algebras, which as well as features for describing systems, come equipped with techniques for comparing systems and proving properties about them. Formal languages such as process algebras do not evolve naturally in the way that human languages do. Instead, they need to be carefully designed.
This fellowship will bring the benefits of the experience of computer science in formal language design to the design of a new process algebra for modelling biological systems, the language BioSPA. One crucial aspect of this new language will be its use of mathematically quantified randomness. In every living thing is a controlled amount of randomness because living things are fuelled by chemical reactions. Systems of chemical reactions evolve stochastically because of the inherent randomness of thermal molecular motion. The BioSPA language will be supported by software tools which will allow biologists to see their models in a number of ways. For example, as a Markov process, a mathematical process which randomly moves between a number of distinct states, or as a series of ordinary differential equations. For either representation it will be possible to trace a possible evolution of the system or to solve the model numerically. The latter could allow a biologist to predict a key value at a specific time, such as the concentration of a particular molecule after ten minutes. A unique feature of the BioSPA language will be that it will also be possible to systematically prove properties of models in the language in order to ensure that they represent the system which the biologists wish to study. This strong, well-designed language will help to ensure the correctness of biological research and the validity of its results, with attendant benefits for medical and pharmaceutical research leading to improvements in the prevention and treatment of illness.

Sethu Vijayakumar appointed IPAB Director

Sethu Vijayakumar will be taking over as Director of IPAB for the 2005-2006 academic year. Thanks and congratulations!